The Task Graph starts with public occupation tasks, attaches research only through recorded matches, and keeps technical exposure separate from observed AI use. The headline uses one named study and one documented calculation, so its meaning stays the same on every role page. Use it to identify activities worth closer investigation, then add the actual workflow, workload, tools, controls, and human context before making a decision.
Headline role score
OpenAI labels each matched O*NET task E0, E1, or E2. E1 means the exposure threshold can be met with a language model alone. E2 means complementary AI-powered software or workflow changes are needed. E0 means the threshold was not met. The headline uses the paper's beta measure: E1 contributes 1, E2 contributes 0.5, and E0 contributes 0.
Example
(2 × 1 + 2 × 0.5 + 1 × 0) ÷ 5 = 60%
In this example, two core tasks are classified E1 and E2 and one supplemental task is E0. Following the OpenAI occupation method, core tasks count twice supplemental tasks. The source mix is 40% E1, 40% E2, and 20% E0. In the beta score, E1 contributes 40 points, E2 contributes 20 points, and E0 contributes none, producing 60%.
Source meanings
Could an LLM or LLM-powered tool cut task time by at least half?
How much task time could a text-only chatbot save?
Was AI use matching this task observed in the source snapshot?
How important, relevant, and frequent is the task in this occupation?
Evidence matching
Used for the OpenAI task classifications behind the headline. The source and The Task Graph must share the O*NET occupation code and task ID.
Used for supporting evidence across O*NET releases after deterministic cleanup of case, punctuation, and spacing. It does not enter the headline percentage.
Used for observed-use data that publishes task text without an occupation key. The page says so explicitly because the match is broader.
Each evidence record stores the source, release, original task key or wording, mapping method, source URL, dataset URL, native label or value, and source-specific limitations. The public page shows the fields needed to understand the result without exposing local file paths or internal processing traces.
Work-time weighting
O*NET publishes task frequency, but frequency is not duration. We therefore do not use it to claim how much of a role's working time is affected. The public score weights core tasks twice supplemental tasks because that is the published OpenAI occupation method; it does not pretend this is a split of working hours.
Personal and organization reports
A person or organization can state how much work time each task takes. That can support a separate workload-weighted view of the same source-backed task classifications. Missing or incomparable workload values must remain missing instead of being guessed. The public occupation baseline always keeps the published core-versus- supplemental weighting and the beta E1/E2/E0 contributions.
Limitations
Inspect further
The source registry lists dataset versions and caveats. Every role page lets you open a task and inspect the result that contributed to the role exposure score.